OK that means your model is mis-specified. Unfortunately I won’t be able to dig into why but I recommend you start with the simplest model you can, get that working on NUTS, and then slowly add back complexity while always making sure NUTS samples well and fast.
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| BNN regression using ADVI | 10 | 1282 | May 14, 2019 | |
| Average Loss in optimization output | 20 | 4788 | July 20, 2018 | |
| ADVI start with initialization | 7 | 2090 | September 23, 2017 | |
| Inf Average loss ADVI in Correlated Topic Model | 9 | 1567 | July 25, 2018 | |
| Infinite loss with ADVI | 1 | 1119 | January 24, 2018 |